No Huddle Offense

"Individual commitment to a group effort-that is what makes a team work, a company work, a society work, a civilization work."

Intent Driven Orchestration

December 12th, 2022 • Comments Off on Intent Driven Orchestration

So let’s start with a bolt statement: the introduction of Microservices/functions and Serverless deployment styles for cloud-native applications has triggered a need to shift the orchestration paradigms towards an intent-driven model.

So what are intents – and what does intent-driven mean? Imagine a restaurant and you order a medium rare steak – the “medium rare” part is the intent declaration. But if we contrast this concept to how orchestration stacks work today – you’d walk into the restaurant, walk straight into the kitchen and you’d say “put the burner on 80% and use that spatula” etc. Essentially declaratively asking for certain amounts of resources/certain way of setup. And obviously, there are a couple of issues with that – you do not necessarily know all the details of the burner. Should it have been set to 80% or 75% maybe? Should it have been 1 core, 500Mb or RAM, sth else? Abstractions and Serverless, anyone?

So why not let app/service owners define what they care about – the objectives of their app/service? For example, “I want P99 latency to be less than 20ms”. That is the “medium rare” intent declaration for an app/service. That is what we’ve been working on here at Intel – and now we’ve released our Intent-Driven Orchestration Planner (Github) for Kubernetes.

Btw.: I shamelessly stole the restaurant metaphor from Kelsey Hightower – for example, check out this podcast. On the P-numbers – again sth that other people have been writing about as well, see Tim Bray‘s blog post on Serverless (part of a Series).

Based on the intents defined by the service owner we want the orchestration stack to handle the rest – just like a good chef. We can do this through scheduling (where/when to place) and planning (how/what to do), to figure out how to set up the stack to make sure the objectives (SLOs) are met.

So why though a planner? The planning component brings sth to the table that the scheduler cannot. It continuously tries to match desired and current objectives of all the workloads. It does this based on data coming from the observability/monitoring stack and tries to reason to enable efficient management. In doing so it can trade-off between various motivations for the stakeholders at play and even take proactive actions if needed – the possibilities for a planner are huge. In the end, the planner can e.g. modify POD specs so the scheduler can make more informed decisions.

Here is an example of that an intent declaration for out Intent Driven Orchestration Planner can look like – essentially requesting that P99 latency should be below 20ms for a target Kubernetes Deployment:

apiVersion: "ido.intel.com/v1alpha1"
kind: Intent
metadata:
  name: my-function-intent
spec:
  targetRef:
    kind: "Deployment"
    name: "default/function-deployment"
  objectives:
    - name: my-function-p99compliance
      value: 20
      measuredBy: default/p99latency

Again the usage of planners is not revolutionary per se, NASA has even flown them to space – and could demonstrate some nice self-healing capabilities – on e.g. Deep Space 1. And just as Deep Space 1 was a tech demonstrator, maybe a quick note: this is all early days for intent-driven orchestration, but we would be very interested in learning what you thinkā€¦

So ultimately, by better understanding the intents of the apps/services instead of just their desires declarative state, orchestrators – thanks to an intent-driven model – can make decisions that will lead to efficiency gains for service and resource owners – sounds like a good thing to me!

AI reasoning & planning

January 4th, 2020 • Comments Off on AI reasoning & planning

With the rise of faster compute hardware and acceleration technologies that drove Deep Learning, it is arguable that the AI winters are over. However Artificial Intelligence (AI) is not all about Neural Networks and Deep Learning in my opinion. Even by just looking at the table of contents of the book “AI – A modern approach” by Russel & Norvig it can easily be seen, that the learning part is only one piece of the puzzle. The topic of reasoning and planning is equally – if not even more – important.

Arguably if you have learned a lot of cool stuff you still need to be able to reason over that gained knowledge to actually benefit from the learned insights. This is especially true, when you want to build autonomic systems. Luckily a lot of progress has been made on the topic of automated planning & reasoning, although they do not necessarily get the same attention as all the neural networks and Deep Learning in general.

To build an autonomous systems it is key to use these kind of techniques which allow for the system to adapt to changes in the context (temporal or spatial changes). I did work a lot on scheduling algorithms in the past to achieve autonomous orchestration, but now believe that planning is an equally important piece. While scheduling tells you where/when to do stuff, planning tells you what/how to do it. The optimal combination of scheduling and planning is hence key for future control planes.

To make this more concrete I spend some time implementing planning algorithms to test concepts. Picture the following, let’s say you have two robot arms. And you just give the control system the goal to move a package from A to B you want the system to itself figure out how to move the arms, to pick the package up & move it from A to B. The following diagram shows this:

(Click to enlarge & animate)

The goal of moving the package from A to B is converted into a plan by looking at the state of the packages which is given by it’s coordinates. By picking up the package and moving it around the state of the package hence changes. The movement of the robot arms is constraint, while the smallest part of the robot arm can move by 1 degree during each time step the bigger parts of the arm can move by 2 & 5 degrees respectively.

Based on this, a state graph can be generated. Where the nodes of the graph define the state of the package (it’s position) and the edges actions that can be performed to alter those states (e.g. move a part of an robot arm, pick & drop package etc.). Obviously not all actions would lead to an optimal solution so the weights on the edges also define how useful this action can be. On top of that, an heuristic can be used that allows the planning algorithm to find it’s goal faster. To figure out the steps needed to move the package from A to B, we need to search this state graph and the “lowest cost” path between start state (package is at location A) and end state (package is at location B) defines the plan (or the steps on what/how to achieve the goal). For the example above, I used D* Lite.

Now that a plan (or series of step) is known we can use traditional scheduling techniques to figure out in which order to perform these. Also note the handover of the package between the robots to move it from A to B this shows – especially in distributed systems – that coordination is key. More on that will follow in a next blog-post.